[PS-2.19] Possible Mechanisms of the Yes Bias: Inquiry from the Data of Human-Robot Interaction Experiment

Oka, N. , Hattori, Y. , Teraoka, K. & Fukada, C.

Kyoto Institute of Technology

Fritzley and Lee (2003) found that 2- and 3-year-olds exhibited a yes bias to simple yes-no questions. Moreover, Okanda and Itakura (2010) showed that 3-year-olds tended to inappropriately say yes to yes-no questions, although they knew the answers to the questions. In our experiment, a robot concurrently learned the names of things and the appropriate responses to three modalities of utterances: telling the name of things, a request for an agreement, and a yes-no question. The experimental result showed that the robot exhibited a yes bias even though it had already learned the correct names of things.
This presentation examines the mechanism of the yes bias of the robot and discusses the implication in the children's yes bias. The robot was equipped with a Memorization Module (M-module) and a Recall & Comparison Module (RC-module) and learned to select an appropriate module according to the modality of utterance. The two possible causes of the bias are:
1) The yes responses by the M-module were incorrectly reinforced because participants tended to ask questions in which they expected the robot to answer 'yes'.
2) The RC-module inappropriately often got negative rewards while name learning was immature.